DALLE-2 vs Stable Diffusion vs Midjourney

➡️ ACCÉDER A chat gpt français en cliquant dessus ! ⬅️

Dally 2, Stable Diffusion et Mid Journey : Quel est le meilleur outil pour créer de l’art à partir de texte ?


Dally 2, Stable Diffusion et Mid Journey sont tous des outils qui permettent de créer de l’art à partir de texte. Chacun de ces outils offre des avantages et des inconvénients différents et offre également des fonctionnalités supplémentaires. Dans ce résumé, nous examinerons les différentes fonctionnalités et les avantages et les inconvénients de chaque outil afin que vous puissiez décider lequel est le meilleur pour vous.

Dally 2

Dally 2 est un outil qui permet de créer de l’art à partir de texte. Il est disponible gratuitement dans certains pays et vous offre 50 crédits gratuits pour votre premier mois et 15 crédits supplémentaires chaque mois. De plus, vous pouvez acheter des crédits supplémentaires pour 15 crédits pour 115 crédits.

Stable Diffusion

Stable Diffusion est un autre outil qui permet de créer de l’art à partir de texte. Il est disponible sous forme open source et peut être téléchargé et installé sur votre ordinateur. Il offre également une version en ligne facile à utiliser avec une version d’essai gratuite.

Mid Journey

Mid Journey est un autre outil qui permet de créer de l’art à partir de texte. Il fonctionne avec un modèle d’abonnement et vous permet de disposer de 200 minutes de temps GPU par mois pour 10 dollars par mois. Vous pouvez également payer 20 dollars supplémentaires par mois pour une instance privée.

Comparaison des outils

Pour comparer les outils, nous allons exécuter le même ensemble de requêtes sur chaque outil et voir quels résultats nous obtenons. Nous allons commencer par Dally 2 et créer une image d’un bar cyberpunk peuplé de cyborgs dans un style anime. Dally 2 a produit quatre résultats intéressants et a bien compris la requête.

Nous allons ensuite passer à Dream Studio Light et créer la même image. Dream Studio Light a produit quatre images intéressantes et a bien compris la requête. Cependant, il n’a pas ajouté de cyborgs.

Enfin, nous allons passer à Mid Journey. Mid Journey est le plus lent des trois outils et nécessite l’utilisation de Discord. Quand nous avons exécuté la requête, il a produit plusieurs images intéressantes et a bien compris la requête. Il a également ajouté des cyborgs.


Dans l’ensemble, Dally 2, Stable Diffusion et Mid Journey sont tous des outils puissants qui peuvent être utilisés pour créer de l’art à partir de texte. Chacun d’entre eux a ses propres avantages et inconvénients, et ils offrent tous des fonctionnalités supplémentaires. Dans l’ensemble, nous avons trouvé que Mid Journey était le meilleur outil, car il a produit les meilleurs résultats et a bien compris la requête.

Source : Gamefromscratch | Date : 2022-10-14 16:31:03 | Durée : 00:17:03

➡️ ACCÉDER A chat gpt français en cliquant dessus ! ⬅️



Support : https://www.patreon.com/gamefromscratch
GameDev News : https://gamefromscratch.com
GameDev Tutorials : https://devga.me
Discord : https://discord.com/invite/R7tUVbD
Twitter : https://twitter.com/gamefromscratch


Stable diffusion does better results but require more prompts

Nicolas Silva A.

midjourney is so ahead of every other ai is kinda disappointing how is only available on discord and every prompt you make is public, also is quite expensive, but i think is worth it considering how the results you get are top tier

Zara T

It's not even close. Midjourney is head and shoulders above all of them


I really hate the fact that most of these are commercialised. From a developer's perspective, I understand it, they need resources to improve the system after all. But as a potential user, I cannot bring myself to pay for subscriptions or even one-time purchases of tools that I don't yet know how to use properly. Heck, I even pirated programs for my 3D studying!

Ryan Pressler

I have a trick with Midjourney that pretty much guarantees something at the quality of a professional srtist

Theodore Gorgovelt



Midjourney Fun fact: even with a free account, you can directly DM the bot your prompts.

Anestis Pneuma

Midjourney released V4 recently (still in beta) which kills the competition by a long shot in my opinion.. I ran the same commands as you did in the video and they all came up significantly better.

Home Cactus

As someone who has used MidJourney quite a bit, there are image parameters that can be added at the end of a prompt. For example add "–v 4" for the version four of the ai. Seems to make realistic images very well and interpret your prompt much better. And more like "–ar x:y" for a custom aspect ratio.


You are not looking at the Midjourney final images. The results are incomplete until you've made at least 1 upscale.


Midjourney released the V.4 update that basically destroys every other ia text to image generator, seriously, its good as hell


DALL E 2 thought they were the best and noone could defeat them and placed a LOAD TON!! of restriction while making it fully paid and locking like its some kind of precious jewel but soon got overtaken and popularity dropped to nothing

midjourney its like an artist which take inspiration from other's artworks and make it in its own style

stable diffusion jack of all trades which soon will be the master and is a work of community together and best thing is its open source

Jose Asse

Hello, first of all thank you for the video, it’s great content! I would only like to point out something about the prices since you talk about it on the beginning.. midjourney’s $30 usd monthly plan offers a “relax mode” which lets you generate prompts a bit slower without using GPU time.. this way I generated over 9,000 images with a single monthly payment.
I think it’s very relevant, since its pretty much an “unlimited” plan that the others don’t offer at this time.

Bartosz Bielecki

SD with inpaiting is super great for synthesized images based on some style you've chosen (and it's not a famous artist).


There are huge ethical problems to using these tools because they are trained from random art on the internet with zero attribution or consent. Some generated images can downright copy a copyrighted work and you have little to no way of knowing it. You also can't copyright them which means your game assets and characters are all in the public domain if you generate them with these AI.

For now I would stay away from using these in commercial projects until the ethical problems are cleared up.


im playing around with waifu diffussion locally.
you can get decent ish results, however, it takes quite a lot more prompts and trial and error than the paid ones.
i havent anything done in novelai, however, from what i can see from other peoples work it is quite impressive.


Thank you for the helpful comparison. Midjouney seemed best for concepts. Also, it seems like a user who knows art or photography jargon would be better able to come up with useful prompts.


Midsummer is Midjourney 2.0.

Ben Verschoor

Egh, I'm leery about the ethics of these AI generators, and this video gets at a lot of reasons why.



oof this is not a great video for a new person to this kind of thing, and pretty bad comparison of the different projects

renato grottesi

You can /imagine in a private message to the midjourney bot, so that your images are not mixed with the other ones from the newbies room

Anirban Basak

Having used all 3 tools compared in this video, I'd like to point out a very important point. Imagine these tools to be artists that speak different languages. Using the same prompt with all of them is like trying to speak to all of them with the same language (most often none of them really understand your inputs). These tools are akin to different artists an agency would have at their disposal for a variety of jobs. For eg. Midjourney shines at abstract concepts, DallE2 is great for photo-realism and also inpainting and outpainting, while Stable Diffusion allows one to be very specific with their prompt.

My suggestion would be to learn the basics of these tools before making a comparison video, as it puts off a lot of newbies from even exploring these tools when they see more failures than successes. Cheers!

The Rising Ape

To be fair, Dreamstudio Lite is NOT Stable Diffusion. It's a really stripped back version that has no where near the generation options that the full version has.


Prompting is like programming in natural language, it's not just writing a woman on a dinosaur, you need to direct it and know how to direct it. I am in no way saying it's hard as programming or that you have full control, there is always that surprise aspect, but you do need some experience. With that said, stable diffusion can do much much better…

insum snoy

I think Midjourney automaticall ychooses a random artist to style your prompt on if you dont specify one yourself, thats why your jets looked so much better.

anti/HUMAN Designs

I've used Dall-e and Stable diffusion, and my conclusion so far is that stable Diffusion is not as versatile and doesn't listen to what you're asking as precisely.
However, it produces nice pictures and is particularly good at producing portraits.
I think Dall-e uses a larger neural network, which would explain why it is "smarter".

And, no, stable diffusion can't do pixel art at all.

Apatsa Wa Basiteni

For artistic or concept art Midjourney and Stable diffusion are good at that, but when it comes to generating 3D textures and photorealistic things DALL-E is on another level; the amount of detail and texture it produces in a photo is just astonishing. Use all of them😁


I've spent some time messing around with all of these and here's my personal experience and findings.

+ Best at photorealism.
+ Excellent at natural language comprehension. You can give it big, long sentences and it generally does a good job at understanding what you want. More so than the others.
– Only available on their servers
– Very censored. Even if you're doing something harmless it often won't let you… it just refuses to accept a lot of prompts like ones that have gun or a celebrity name in them e.t.c

+ Best at art
+ Very creative. You might find this annoying but if you're using it for concepting or just making cool images MidJourney will usually produce much more creative/artistic results
-/+ Has a pretty distinct style of its own. You can absolutely tweak the output and get different styles, but it does kind of have its own identity and that can be both a pro or con, depends.
– A little slow compared to the others
– Only available on their discord
– Somewhat censored. Not as much as DALL.E but something worth considering

Stable Diffusion
+ Pretty good at both photo-realism and art. I'd say a little better at art than realism but balanced overall
+ Fast and light weight, part of why you can run it on normal consumer hardware
+ Plenty of community/open-source forks. Lots to choose from
+ Can run locally on your own hardware. Great for privacy and not being censored. You can run it on pretty normal consumer hardware. I use it on an 8GB GTX 1070 without trouble
+ You can train your own models and yourself or friends and family to make cool images with your likeness. Can also train custom art styles, great for artists.
– /+ Has an official hosted version like the others. Good for beginners or people trying it out. Same cons as the others in terms of privacy and censorship (still less than DALL.E)
– Not very good at natural language comprehension. Your prompts will require more engineering compared to the other two
– Running locally is pretty easy depending on the version you use, but it's still a small difficulty jump compared to the web versions (it's really not that hard tho)


tell the people that kitbash3d is giving minarva assets free right now.


Stable Diffusion has a major advantage of being able to run it locally on as little as 4 GB of VRAM, I can generate 512×512 images on my old GTX 960 in less than a minute, with all content limitations disabled (nsfw, irl people, etc.). But I can go as high as ~850×850.

Also the ability to fine-tune the model, there's already Waifu Diffusion for anime, or NovelAI's finetuned SD version for anime. Finetune for furries, hell, there are finetune for just VTubers.

The major limitation of SD is the pixelart, Dall-E definitely wins there, but maybe it could be finetuned?


What is this tech actually doing though? Scraping the internet for images? Compositing images or building the pictures pixel by pixel


There is also the WOMBO Dream for the fourth corner. It behaves slightly differently with the style choice, but it's an honest competitor (and I like it).


Half of it is the prompt? Sure, but in my experience the other half is just trying many many times.

Todd Zircher

The use of commas to break sentences into key phrases is important for the AI. Generally like this:
primary subject, details, details, rendering options, art style
sad woman, beach at sunset, burning sky, photorealistic, shadows.


Microsoft just announced Dall-E 2 will be added to 365 (office) and to Bing search. This makes

A Disibio

The problem with Dalle-2 is it is only as creative as you are.


does anyone know if these tools are being archived at each version release? it would be cool to come back to these in a few decades 🙂

Eric Lefebvre

MidJourney has a new engine you can render with by using –test in your prompt. It generates much better images in many cases.

Maya Anonymous

"create art from text"
no, create art from ripped off artists that you can search for


Really great work as always. Good idea to do a side by side comparison. these types of networks can also be used for Textures! "Texture brickwall pixel art" and "brickwall texture in normal map style" because a normal map is a just a expressionism art style, to a style transfer gan.


It’s too early. There are a few things they are good at (landscapes, face portraits with some caveats) but anything resembling a specific composition or a character doing a specific action always ends in a spectacular failure. Even when you get something usable, the details are pure chaos and doesn’t look right. They show incredible future potential but that’s it.

Eric Lefebvre

I'm working on a boardgame and creating my art with MidJounrey but looking to move to Stability Diffusion. Honestly, I find Midjourney to be way more interesting in terms of results. I'm running SD on my gaming rig. A ryzen 7 3700X, 32GB of Ram and a RTX 3070 locally and it works great.

Paulo Rodriguez

stable diffusion all the way! it has evolved so much by the community because open source , it's just beautiful.
i'm using textual inversion to create more concept art for some of my characters. plus no BS censorship.


Stable diffusion is actually really powerful and it didn't get showcased super well here. It does take some experience to know how to write the prompts. I think they tend to understand better when you have a bunch of short, comma separated phrases.

Ivan Isakovic

Prompt craft is a thing. The way to get the AI to give expected and good results in a desired style. Respectfully, the way you did it is not ideal to show capabilities of each AI.

Veuillez vous connecter pour laisser un commentaire.